Materialize global addresses via movt/movw pair, this is always better
than doing the same via constpool:
1. Load from constpool costs 3 cycles on A9, movt/movw pair - just 2.
2. Load from constpool might stall up to 300 cycles due to cache miss.
3. Movt/movw does not use load/store unit.
4. Less constpool entries => better compiler performance.
This is only enabled on ELF systems, since darwin does not have needed
relocations (yet).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@89720 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp b/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
index daffc5e..692bb19 100644
--- a/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
+++ b/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
@@ -330,6 +330,8 @@
void ARMAsmPrinter::printOperand(const MachineInstr *MI, int OpNum,
const char *Modifier) {
const MachineOperand &MO = MI->getOperand(OpNum);
+ unsigned TF = MO.getTargetFlags();
+
switch (MO.getType()) {
default:
assert(0 && "<unknown operand type>");
@@ -356,12 +358,12 @@
case MachineOperand::MO_Immediate: {
int64_t Imm = MO.getImm();
O << '#';
- if (Modifier) {
- if (strcmp(Modifier, "lo16") == 0)
- O << ":lower16:";
- else if (strcmp(Modifier, "hi16") == 0)
- O << ":upper16:";
- }
+ if ((Modifier && strcmp(Modifier, "lo16") == 0) ||
+ (TF & ARMII::MO_LO16))
+ O << ":lower16:";
+ else if ((Modifier && strcmp(Modifier, "hi16") == 0) ||
+ (TF & ARMII::MO_HI16))
+ O << ":upper16:";
O << Imm;
break;
}
@@ -371,6 +373,13 @@
case MachineOperand::MO_GlobalAddress: {
bool isCallOp = Modifier && !strcmp(Modifier, "call");
GlobalValue *GV = MO.getGlobal();
+
+ if ((Modifier && strcmp(Modifier, "lo16") == 0) ||
+ (TF & ARMII::MO_LO16))
+ O << ":lower16:";
+ else if ((Modifier && strcmp(Modifier, "hi16") == 0) ||
+ (TF & ARMII::MO_HI16))
+ O << ":upper16:";
O << Mang->getMangledName(GV);
printOffset(MO.getOffset());